A multitude of algorithms have already been known whose goal is to identify characters in an image, for example on a scanned page. In the method mentioned, templates describing known characters are directly compared to the image, for example. However, such a comparison of a template with an image is extremely costly in terms of computing expenditure. In addition, most known methods have a low level of reliability if different fonts are to be detected. Also, the detection of handwriting is particularly problematic since, as is known, there are huge differences between the different persons' handwriting.